Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Pays Bas → 2020 → Remote sensing based pre-season yellow rust early warning in Ethiopia.

University of Twente - International Institute for Geo-Information Science and Earth Observation (ITC) 2020

Remote sensing based pre-season yellow rust early warning in Ethiopia.

Endo, Chinatsu

Titre : Remote sensing based pre-season yellow rust early warning in Ethiopia.

Auteur : Endo, Chinatsu

Etablissement de soutenance : University of Twente - International Institute for Geo-Information Science and Earth Observation (ITC)

Grade : Master of Science in Geo-Information Science and Earth Observation 2020

Résumé
Yellow rust (Puccinia striiformis f. sp. Tritici) is a crop disease of wheat that regularly causes yield loss in Ethiopia. The disease has significant consequences for the country’s crop production, food security, health, and socioeconomic well-being. Anticipating yellow rust epidemics can help better manage them and mitigate their adverse impacts. This study explores the potential of remote sensing-based early prediction of yellow rust in the Oromia region in Ethiopia. The research focuses on modeling the incidence of yellow rust among young wheat in the region by looking at unique environmental conditions that enable off-season survival of the rust pathogen. Tiller and boot-level yellow rust incidence data from 2016-2018 in Oromia was analyzed together with the environmental variables generated through AgERA5 (temperature), CHIRPS (precipitation), ProbaV-NDVI, and SRTM-DEM (terrain characteristics). Univariate Area Under ROC Curve analysis and Classification Tree analysis were used to understand the influential environmental variables and filter those with high relevance to the early-stage rust infection. Subsequently, General Additive Model and Boosted Regression Tree were applied to fit and test the early warning models and their prediction capacity. The models were built for three data sets : data with all available observations ; tiller-level observations ; and data that share the same climate zone. As a result, the climate zone-based GAM model performed at a 78% accuracy level with Kappa 0.44 (moderate). The tiller-only GAM model performed at a 72% accuracy level with Kappa 0.44 (moderate). The all-observation BRT model had a 71% accuracy level with Kappa 0.34 (fair agreement). Rain characteristics served as particularly strong predictors in these models. Especially, excessive rain had a strong relationship with a lower probability of yellow rust cases among young wheat. The models also suggest that terrain characteristics serve as the static environmental conditions that expose certain locations to the disease. The study demonstrated the potential of yellow rust early warning solely based on remote sensing. The models could be further tested with a larger volume of data set to confirm the strength. Consideration of the probability of varying rust severity (low, moderate, high) and types of wheat cultivars would further add value to the models. Lastly, additional field and laboratory-based knowledge on the off-season rust survival would be a vital step towards a more accurate configuration of early warning models.

Présentation

Version intégrale (ITC)

Page publiée le 22 décembre 2021